def flipAndSaveToRAS(filename):
#Recover the image object
imageObj = nib.load(filename)
#Get the current orientation
CurrentOrientation = nib.aff2axcodes(imageObj.affine)
print("The current orientation is : ", CurrentOrientation)
#Check if the current orientation is already RAS+
if CurrentOrientation == ('R', 'A', 'S'):
print(
"Image already recorded into the RAS+ orientation, nothing to do")
return filename
else:
#Flip the image to RAS
flippedImage = nib.as_closest_canonical(imageObj)
##Check the new orientation
NewOrientation = nib.aff2axcodes(flippedImage.affine)
#Set Qcode to 1 that the Qform matrix can be used into the further processing
flippedImage.header['qform_code'] = 1
#Save the flipped image
nib.save(flippedImage, RASFile)
print("The new orientation is now : ", NewOrientation)
return RASFile
def reorient_images(self):
if (self.reorient_flag):
print(
"Reorient flag is set to true, Hence reorienting both images to Right Anterior Superior"
)
canonical_img_1 = nb.as_closest_canonical(self.orig_nii_stationary)
print(" ============= ============== ===================")
print("orientation changed t1 affine: {}".format(
canonical_img_1.affine))
print(" ============= ============== ===================")
print("orientation changed t1 : {}".format(
nb.aff2axcodes(canonical_img_1.affine)))
print(" ============= ============== ===================")
canonical_img_2 = nb.as_closest_canonical(self.orig_nii_moving)
print(" ============= ============== ===================")
print("orientation changed t2 affine: {}".format(
canonical_img_2.affine))
print(" ============= ============== ===================")
print("orientation changed t1 : {}".format(
nb.aff2axcodes(canonical_img_2.affine)))
print(" ============= ============== ===================")
self.canonical_img_1 = canonical_img_1
self.canonical_img_2 = canonical_img_2
return self.canonical_img_1, self.canonical_img_2
else:
print(" ============= ============== ===================")
print("Not reorienting the images as reorient flag is false")
print(" ============= ============== ===================")
self.canonical_img_1 = orig_nii_stationary
self.canonical_img_2 = orig_nii_moving
return self.canonical_img_1, self.canonical_img_2
def _reorient_image(img, *, target_img=None, orientation=None):
"""
Coerce an image to a target orientation.
.. note::
Only RAS -> LAS conversion is currently supported
Parameters
----------
img : :obj:`SpatialImage`
image to be reoriented
target_img : :obj:`SpatialImage`, optional
target in desired orientation
orientation : :obj:`str` or :obj:`tuple`, optional
desired orientation, if no target image is provided
.. testsetup::
>>> img = nb.load(Path(test_data) / 'testRobustMNINormalizationRPTMovingWarpedImage.nii.gz')
>>> las_img = img.as_reoriented([[0, -1], [1, 1], [2, 1]])
Examples
--------
>>> nimg = _reorient_image(img, target_img=img)
>>> nb.aff2axcodes(nimg.affine)
('R', 'A', 'S')
>>> nimg = _reorient_image(img, target_img=las_img)
>>> nb.aff2axcodes(nimg.affine)
('L', 'A', 'S')
>>> nimg = _reorient_image(img, orientation='LAS')
>>> nb.aff2axcodes(nimg.affine)
('L', 'A', 'S')
>>> _reorient_image(img, orientation='LPI')
Traceback (most recent call last):
...
NotImplementedError: Cannot reorient ...
>>> _reorient_image(img)
Traceback (most recent call last):
...
RuntimeError: No orientation ...
"""
orient0 = nb.aff2axcodes(img.affine)
if target_img is not None:
orient1 = nb.aff2axcodes(target_img.affine)
elif orientation is not None:
orient1 = tuple(orientation)
else:
raise RuntimeError("No orientation to reorient to!")
if orient0 == orient1: # already in desired orientation
return img
elif orient0 == tuple("RAS") and orient1 == tuple("LAS"): # RAS -> LAS
return img.as_reoriented([[0, -1], [1, 1], [2, 1]])
else:
raise NotImplementedError("Cannot reorient {0} to {1}.".format(
orient0, orient1))
def _run_interface(self, runtime):
dsi_studio_file = self.inputs.dsi_studio_nifti
new_file = fname_presuffix(dsi_studio_file,
suffix="fixhdr",
newpath=runtime.cwd)
dsi_img = nb.load(dsi_studio_file)
correct_img = nb.load(self.inputs.correct_header_nifti)
new_axcodes = nb.aff2axcodes(correct_img.affine)
input_axcodes = nb.aff2axcodes(dsi_img.affine)
# Is the input image oriented how we want?
if not input_axcodes == new_axcodes:
# Re-orient
input_orientation = nb.orientations.axcodes2ornt(input_axcodes)
desired_orientation = nb.orientations.axcodes2ornt(new_axcodes)
transform_orientation = nb.orientations.ornt_transform(
input_orientation, desired_orientation)
reoriented_img = dsi_img.as_reoriented(transform_orientation)
else:
reoriented_img = dsi_img
# No matter what, still use the correct affine
nb.Nifti1Image(reoriented_img.get_data(),
correct_img.affine).to_filename(new_file)
self._results['out_file'] = new_file
return runtime
def read_input_images(self):
orig_nii_stationary = nb.load(self.stationary_image_file_path)
orig_nii_moving = nb.load(self.moving_image_file_path)
orig_nii_stationary_voxel_dim = orig_nii_stationary.header["pixdim"][1:4]
orig_nii_moving_voxel_dim = orig_nii_moving.header["pixdim"][1:4]
orig_nii_stationary_centre = [float(orig_nii_stationary.header["qoffset_x"]), float(orig_nii_stationary.header["qoffset_y"]), float(orig_nii_stationary.header["qoffset_z"])]
orig_nii_moving_centre = [float(orig_nii_moving.header["qoffset_x"]), float(orig_nii_moving.header["qoffset_y"]), float(orig_nii_moving.header["qoffset_z"])]
print(" ============= ============== ===================")
print("Image 1 voxel resolution before resampling: {}".format(orig_nii_stationary_voxel_dim))
print(" ============= ============== ===================")
print("Image 2 voxel resolution before resampling: {}".format(orig_nii_moving_voxel_dim))
print(" ============= ============== ===================")
print("Image 1 centre before resampling: {}".format(orig_nii_stationary_centre))
print(" ============= ============== ===================")
print("Image 2 centre before resampling: {}".format(orig_nii_moving_centre))
print(" ============= ============== ===================")
print("original t1 affine: {}".format(orig_nii_stationary.affine))
print(" ============= ============== ===================")
print("original t2 affine: {}".format(orig_nii_moving.affine))
print(" ============= ============== ===================")
print("original t1 Orientation: {}".format(nb.aff2axcodes(orig_nii_stationary.affine)))
print(" ============= ============== ===================")
print("original t2 Orientation: {}".format(nb.aff2axcodes(orig_nii_moving.affine)))
print(" ============= ============== ===================")
self.orig_nii_stationary = orig_nii_stationary
self.orig_nii_moving = orig_nii_moving
return self.orig_nii_stationary, self.orig_nii_moving;
def test_MNI_reorient():
# load up the image with the incorrect orientation
bad_img = nib.load(bad_orientation_file)
# get the affine information from the image
bad_affine = bad_img.affine
# get the orientation code from the affine information
bad_orientation = nib.aff2axcodes(bad_affine)
# run the reorientation on the file with the bad orientation to create
# a file with the correct orientation
#if isfile(out_file_name):
# os.remove(out_file_name)
# MNI_reorient(bad_orientation_file, out_file_name)
# load up the image data of the out file created in the previous step
out_file_img = nib.load(out_file_name)
# get the affine information from the out file image
out_file_affine = out_file_img.affine
# get the orientation code from the affine information
out_file_orientation = nib.aff2axcodes(out_file_affine)
# assert that the original image with the incorrect orientation actually does
# have a different orientation than what is desired
assert (bad_orientation != MNI_axis_codes)
# assert that the newly created out file has had its orientation swapped
# to the desired orientation
assert (out_file_orientation == MNI_axis_codes)
def getAxesForTransform(startingDicomFile, cfg):
""" Load one example file """
nifti_object = nib.load(cfg.ref_BOLD)
target_orientation = nib.aff2axcodes(nifti_object.affine)
dicom_object = getLocalDicomData(cfg, startingDicomFile)
dicom_object = dicomreaders.mosaic_to_nii(dicom_object)
dicom_orientation = nib.aff2axcodes(dicom_object.affine)
return target_orientation, dicom_orientation # from here you can save and load it so getTransform is hard coded --you only need to run this once
def get_ornt(self):
"""Returns current orientation based on the affine and coordinate system"""
if self.coord_sys == 'nib':
ornt_tup = aff2axcodes(self.affine)
elif self.coord_sys == 'itk':
ornt_tup = inv_axcodes(aff2axcodes(convert_affine(self.affine)))
else:
raise ValueError(f'Invalid coord_sys: "{self.coord_sys}"')
ornt_str = ''.join(ornt_tup)
return ornt_str
def print_orien(example_file):
img = nib.load(example_file)
# Here is the affine (to two digits decimal precision):
np.set_printoptions(precision=2, suppress=True)
# print(f"img.affine={img.affine}")
# What are the orientations of the voxel axes here?
# Nibabel has a routine to tell you, called aff2axcodes.
orientation = nib.aff2axcodes(img.affine)
print(f"orientation of {example_file} = {orientation}")
return nib.aff2axcodes(img.affine)
def main():
parser = _build_args_parser()
args = parser.parse_args()
assert_inputs_exist(parser, args.input, args.ref)
assert_outputs_exist(parser, args, args.output)
if args.enforce_dimensions and not args.ref:
parser.error("Cannot enforce dimensions without a reference image")
if args.verbose:
logging.basicConfig(level=logging.DEBUG)
logging.debug('Loading Raw data from %s', args.input)
img = nib.load(args.input)
data = img.get_data()
affine = img.get_affine()
original_zooms = img.get_header().get_zooms()[:3]
if args.ref:
ref_img = nib.load(args.ref)
new_zooms = ref_img.header.get_zooms()[:3]
elif args.resolution:
new_zooms = [args.resolution] * 3
elif args.iso_min:
min_zoom = min(original_zooms)
new_zooms = (min_zoom, min_zoom, min_zoom)
logging.debug('Data shape: %s', data.shape)
logging.debug('Data affine: %s', affine)
logging.debug('Data affine setup: %s', nib.aff2axcodes(affine))
logging.debug('Resampling data to %s with mode %s', new_zooms, args.interp)
data2, affine2 = reslice(data, affine, original_zooms, new_zooms,
interp_code_to_order(args.interp))
logging.debug('Resampled data shape: %s', data2.shape)
logging.debug('Resampled data affine: %s', affine2)
logging.debug('Resampled data affine setup: %s', nib.aff2axcodes(affine2))
logging.debug('Saving resampled data to %s', args.output)
if args.enforce_dimensions:
computed_dims = data2.shape
ref_dims = ref_img.shape[:3]
if computed_dims != ref_dims:
fix_dim_volume = np.zeros(ref_dims)
x_dim = min(computed_dims[0], ref_dims[0])
y_dim = min(computed_dims[1], ref_dims[1])
z_dim = min(computed_dims[2], ref_dims[2])
fix_dim_volume[:x_dim, :y_dim, :z_dim] = \
data2[:x_dim, :y_dim, :z_dim]
data2 = fix_dim_volume
nib.save(nib.Nifti1Image(data2, affine2), args.output)
def reorient_image(input_path):
input_image = nib.load(input_path)
print(nib.aff2axcodes(input_image.affine))
output_image = nib.as_closest_canonical(input_image)
output_path = input_path.split('.')[0] + '_reoriented.nii.gz'
print(output_path)
print(nib.aff2axcodes(output_image.affine))
nib.save(output_image, output_path)
def _resample(mask, target_affine, target_shape):
if target_affine is not None and target_shape is not None:
mask = image.resample_img(mask, target_affine=target_affine, target_shape=target_shape, interpolation='nearest')
# check orientations
orient_data = ''.join(nib.aff2axcodes(target_affine))
orient_roi = ''.join(nib.aff2axcodes(mask.affine))
if not orient_roi == orient_data:
msg = 'Orientation of mask and data are not the same: ' + \
orient_roi + ' (mask) vs. ' + orient_data + ' (data)'
logger.error(msg)
raise ValueError(msg)
return mask
def test_orntd_2d(self):
data = {
"seg": np.ones((2, 1, 3)),
"img": np.ones((2, 1, 3)),
"seg.affine": np.eye(4),
"img.affine": np.eye(4)
}
ornt = Orientationd(keys=("img", "seg"), axcodes="PLI")
res = ornt(data)
np.testing.assert_allclose(res["img"].shape, (2, 3, 1))
code = nib.aff2axcodes(res["seg.affine"], ornt.ornt_transform.labels)
self.assertEqual(code, ("P", "L", "S"))
code = nib.aff2axcodes(res["img.affine"], ornt.ornt_transform.labels)
self.assertEqual(code, ("P", "L", "S"))
def test_orntd_1d(self):
data = {
"seg": np.ones((2, 3)),
"img": np.ones((2, 3)),
"seg_meta_dict": {"affine": np.eye(4)},
"img_meta_dict": {"affine": np.eye(4)},
}
ornt = Orientationd(keys=("img", "seg"), axcodes="L")
res = ornt(data)
np.testing.assert_allclose(res["img"].shape, (2, 3))
code = nib.aff2axcodes(res["seg_meta_dict"]["affine"], ornt.ornt_transform.labels)
self.assertEqual(code, ("L", "A", "S"))
code = nib.aff2axcodes(res["img_meta_dict"]["affine"], ornt.ornt_transform.labels)
self.assertEqual(code, ("L", "A", "S"))
def test_orntd_1d(self):
data = {
'seg': np.ones((2, 3)),
'img': np.ones((2, 3)),
'seg.affine': np.eye(4),
'img.affine': np.eye(4)
}
ornt = Orientationd(keys=('img', 'seg'), axcodes='L')
res = ornt(data)
np.testing.assert_allclose(res['img'].shape, (2, 3))
code = nib.aff2axcodes(res['seg.affine'], ornt.ornt_transform.labels)
self.assertEqual(code, ('L', 'A', 'S'))
code = nib.aff2axcodes(res['img.affine'], ornt.ornt_transform.labels)
self.assertEqual(code, ('L', 'A', 'S'))
def test_orntd_canonical(self):
data = {
"seg": np.ones((2, 1, 2, 3)),
"img": np.ones((2, 1, 2, 3)),
"seg.affine": np.eye(4),
"img.affine": np.eye(4),
}
ornt = Orientationd(keys=("img", "seg"), as_closest_canonical=True)
res = ornt(data)
np.testing.assert_allclose(res["img"].shape, (2, 1, 2, 3))
np.testing.assert_allclose(res["seg"].shape, (2, 1, 2, 3))
code = nib.aff2axcodes(res["seg.affine"], ornt.ornt_transform.labels)
self.assertEqual(code, ("R", "A", "S"))
code = nib.aff2axcodes(res["img.affine"], ornt.ornt_transform.labels)
self.assertEqual(code, ("R", "A", "S"))
def resampleNiftifile(self, ):
for root, dirs, files in os.walk(self.resampling_folder):
print()
print("========= ============ =========")
for fl in files:
if (fl.endswith(".nii.gz")):
basename = fl[0:-7]
img_nb = nb.load(os.path.join(self.resampling_folder, fl))
img_np = img_nb.dataobj
target_shape = np.array((256, 256, 256))
new_resolution = [
0.5,
] * 3
new_affine = np.zeros((4, 4))
new_affine[:3, :3] = np.diag(new_resolution)
new_affine[:3, 3] = target_shape * new_resolution / 2. * -1
new_affine[3, 3] = 1.0
new_affine
interim_nb = nl.image.resample_img(
img_nb,
target_affine=new_affine,
target_shape=target_shape,
interpolation='nearest')
print("======== Before ==========")
print("Shape and max intensity of img is {} and {}".format(
img_np.shape, np.max(img_np)))
print("Voxel resolution: {}".format(
img_nb.header["pixdim"][1:4]))
print("Orientation: {}".format(
nb.aff2axcodes(img_nb.affine)))
print("Affine: {}".format(img_nb.affine))
interim_nb.to_filename(
os.path.join(self.resampling_folder,
basename + "_upsampled.nii.gz"))
tmp_nb = nb.load(
os.path.join(self.resampling_folder,
basename + "_upsampled.nii.gz"))
tmp_np = tmp_nb.dataobj
print("========= After ===========")
print()
print("Shape and max intensity of img is {} and {}".format(
tmp_np.shape, np.max(tmp_np)))
print("Voxel resolution: {}".format(
tmp_nb.header["pixdim"][1:4]))
print("Orientation: {}".format(
nb.aff2axcodes(tmp_nb.affine)))
print("Affine: {}".format(tmp_nb.affine))
print()
def load_nifti(filepath):
"""Load a Nifti file.
# Arguments
filepath: The file path of the Nifti image to load.
# Returns
A four dimensional array where the last dimension are the color
channels (rgb). If the Nifti only contains gray values then all
channels have the same value for one voxel."""
img = nib.load(filepath)
img_data = img.get_data()
# make sure volume orientation is LPS (DICOM default)
orientation = ''.join(nib.aff2axcodes(img.affine))
if not orientation == 'LPS':
if orientation == 'LAS':
img_data = np.flip(img_data, 1)
elif orientation == 'RAS':
img_data = np.flip(img_data, 0)
img_data = np.flip(img_data, 1)
else:
raise ValueError('Unsupported orientation of Nifti file: ' +
orientation)
# add a channels dimension (if not already present)
if len(img_data.shape) < 4:
img_data.shape += (1,)
return img_data
def orientation(self):
"""tuple[str]: Image orientation in standard orientation format.
See orientation.py for more information on conventions.
"""
nib_orientation = nib.aff2axcodes(self._affine)
return stdo.orientation_nib_to_standard(nib_orientation)
def read(path='',
b_reorient=False,
orientation=(('R', 'L'), ('P', 'A'), ('I', 'S'))):
"""
Read the nifti file and switch to a given orientation
orientation defaults to std LAS (radiological) - RAS (neurological)
"""
if not os.path.isfile(path):
raise ValueError('Provided path is not a valid file')
image_nii = nib.load(path)
if b_reorient:
# switch to given orientation (http://nipy.org/nibabel/image_orientation.html)
axcodes = nib.aff2axcodes(image_nii.affine)
orientations = nib.orientations.axcodes2ornt(axcodes, orientation)
image = image_nii.get_data()
image = nib.apply_orientation(image, orientations)
header = image_nii.header
img_shape = image.shape
else:
image = image_nii.get_data()
img_shape = image.shape
header = image_nii.header
print(f'preprocessed data shape={img_shape}')
return image, header, img_shape
def _format_volume_to_header(volume: MedicalVolume) -> MedicalVolume:
"""Reformats the volume according to its header.
Args:
volume (MedicalVolume): The volume to reformat.
Must be 3D and have headers of shape (1, 1, volume.shape[2]).
Returns:
MedicalVolume: The reformatted volume.
"""
headers = volume.headers()
assert headers.shape == (1, 1, volume.shape[2])
affine = to_RAS_affine(headers.flatten())
orientation = stdo.orientation_nib_to_standard(nib.aff2axcodes(affine))
# Currently do not support mismatch in scanner_origin.
if tuple(affine[:3, 3]) != volume.scanner_origin:
raise ValueError(
"Scanner origin mismatch. "
"Currently we do not handle mismatch in scanner origin "
"(i.e. cannot flip across axis)")
volume = volume.reformat(orientation)
assert volume.headers().shape == (1, 1, volume.shape[2])
return volume
def _get_mgdm_orientation(affine, mgdm):
'''
Transforms nibabel affine information into
orientation and slice order that MGDM understands
'''
orientation = nb.aff2axcodes(affine)
# set mgdm slice order
if orientation[-1] == "I" or orientation[-1] == "S":
sliceorder = mgdm.AXIAL
elif orientation[-1] == "L" or orientation[-1] == "R":
sliceorder = mgdm.SAGITTAL
else:
sliceorder = mgdm.CORONAL
# set mgdm orientations
if "L" in orientation:
LR = mgdm.R2L
elif "R" in orientation:
LR = mgdm.L2R # flipLR = True
if "A" in orientation:
AP = mgdm.P2A # flipAP = True
elif "P" in orientation:
AP = mgdm.A2P
if "I" in orientation:
IS = mgdm.S2I # flipIS = True
elif "S" in orientation:
IS = mgdm.I2S
return sliceorder, LR, AP, IS
def read_nifti_series(filename):
proxy_img = nib.load(filename)
# less efficent get image data into memory all at once
# image_data = proxy_img.get_fdata()
hdr = proxy_img.header
image_shape = hdr.get_data_shape()
image_dim = len(image_shape)
num_images = 1
if image_dim >= 3:
num_images = image_shape[2]
(m, b) = hdr.get_slope_inter()
axcodes = nib.aff2axcodes(proxy_img.affine)
# TODO-- There does not seem to be a good NIfti method to obtain the input volume's labels?
if ((axcodes != ('R', 'A', 'S')) and (axcodes != ('L', 'A', 'S'))
and (axcodes != ('L', 'P', 'S'))):
print(
"Input NIfti series is in unsupported orientation. Please convert to RAS, LAS, or LPS orientation:"
+ filename)
sys.exit(1)
# specifiy LPS for DICOM
# https://nipy.org/nibabel/dicom/dicom_orientation.html, if we want to apply the formula above to array indices in pixel_array, we first have to apply a column / row flip to the indices.
codes = ('L', 'P', 'S')
labels = (('A', 'P'), ('R', 'L'), ('I', 'S'))
orients = orientations.axcodes2ornt(codes, labels)
img_reorient = proxy_img.as_reoriented(orients)
hdr = img_reorient.header
# We reset m and b here ourselves for downstream rescale/slope
b = 0
m = 1
return img_reorient, hdr, num_images, b, m, axcodes
def read_nii_from_path(config, path=None):
"""
Read nii file from the path
:param config: type dict: config parameter
:param path: path to nii.file
:return: img_arr : type ndarray: 3D array from nii file
:return: img.header: type nibabel.nifti1.Nifti1Header: header of nifti image
:return: img.affine: type ndarray, affine info of nifti image
"""
print(path)
img = nib.load(path)
axcodes = tuple(nib.aff2axcodes(img.affine))
img_arr = img.get_fdata()
if config['new_orientation']:
img_arr = nifti_reorientation(
np.array(img_arr), axcodes,
tuple(config['new_orientation'])) #('P','L','S')
if config['rescale']:
slope = config['rescale'] / (np.max(img_arr) + 1e-16)
img_arr = img_arr / (np.max(img_arr) + 1e-16) * config['rescale']
img.header.set_slope_inter(slope, inter=0)
elif config['scale']:
print(config['scale'])
img_arr = img_arr * config['scale']
else:
pass
return img_arr, img.header, img.affine
def _resample(mask, target_affine, target_shape):
if target_affine is not None and target_shape is not None:
mask = image.resample_img(
mask,
target_affine=target_affine,
target_shape=target_shape,
interpolation="nearest",
)
# check orientations
orient_data = "".join(nib.aff2axcodes(target_affine))
orient_roi = "".join(nib.aff2axcodes(mask.affine))
if not orient_roi == orient_data:
logger.error(
"Orientation of mask and data are not the same: " +
orient_roi + " (mask) vs. " + orient_data + " (data)")
return mask
def vtk_to_trk_converter(tract_in, tract_out, reference_volume=None):
if not tract_in.endswith('vtk'):
raise ValueError("Sorry, we only work with vtk files as input")
if not tract_out.endswith('trk'):
raise ValueError("Sorry, we only work with trk files as output")
tractography = tractography_from_vtk_file(tract_in)
# ADD metadata to the TRK format
hdr_dict = None
if not (reference_volume is None):
ref = nibabel.load(reference_volume)
hdr_dict = {'dimensions': ref.shape[:3],
'voxel_sizes': ref.header.get_zooms()[:3],
'voxel_to_rasmm': ref.affine,
'voxel_order': "".join(nibabel.aff2axcodes(ref.affine))}
tract = nibabel.streamlines.Tractogram(tractography.tracts(),
affine_to_rasmm=np.eye(4))
trk_file = nibabel.streamlines.TrkFile(tract, hdr_dict)
trk_file.save(tract_out)
def orientation(self):
"""
Get the closest orientation in standard orientation coordinates
:return: a tuple of standard orientation coordinates (see orientation.py for more information on format)
"""
nib_orientation = nib.aff2axcodes(self._affine)
return stdo.orientation_nib_to_standard(nib_orientation)
def test_orntd_no_metadata(self):
data = {"seg": np.ones((2, 1, 2, 3))}
ornt = Orientationd(keys="seg", axcodes="RAS")
res = ornt(data)
np.testing.assert_allclose(res["seg"].shape, (2, 1, 2, 3))
code = nib.aff2axcodes(res["seg_meta_dict"]["affine"], ornt.ornt_transform.labels)
self.assertEqual(code, ("R", "A", "S"))
def __call__(self, data):
"""
:param data: Data dictionary to be processed by this transform
:type data: dict
:return: Updated data dictionary
:rtype: dict
"""
for field in self.fields:
complete_file_path = os.path.join(self.data_dir, data[field])
assert complete_file_path[0:5] == 's3://'
filename = get_file_from_s3(self.s3_client, complete_file_path, self.cache)
img = nib.load(filename)
if self.canonical:
img = nib.as_closest_canonical(img)
data[field] = img
data[field + '_affines'] = img.affine
data[field + '_orientations'] = nib.aff2axcodes(img.affine)
return data
def split_seg_sides(in_bin_seg_file, out_seg_file):
"""
Split segmentation into Right/Left
:param in_bin_seg_file: input binary segmentation
:param out_seg_file: output segmentation with both sides
"""
in_bin_seg = nib.load(in_bin_seg_file)
mid = int(in_bin_seg.shape[0] / 2)
out_seg = in_bin_seg.get_data().copy()
seg_ort = nib.aff2axcodes(in_bin_seg.affine)
r_orient_nii = ('R', 'A', 'S')
l_orient_nii = ('L', 'A', 'S')
if seg_ort == l_orient_nii:
# new = in_bin_seg.get_data()[mid:-1, :, :]
# new[new == 1] = 2
# out_seg[mid:-1, :, :] = 2
out_seg[0:mid, :, :] = 0
out_seg = out_seg + in_bin_seg.get_data()
elif seg_ort == r_orient_nii:
# new = in_bin_seg.get_data()[0:mid, :, :]
# new[new == 1] = 2
# out_seg[0:mid, :, :] = 2
out_seg[mid:-1, :, :] = 0
out_seg = out_seg + in_bin_seg.get_data()
print(out_seg.max())
out_seg_nii = nib.Nifti1Image(out_seg, in_bin_seg.affine)
nib.save(out_seg_nii, out_seg_file)
def spikes_mask(in_file, in_mask=None, out_file=None):
"""
Utility function to calculate a mask in which check
for :abbr:`EM (electromagnetic)` spikes.
"""
import os.path as op
import nibabel as nb
import numpy as np
from nilearn.image import mean_img
from nilearn.plotting import plot_roi
from scipy import ndimage as nd
if out_file is None:
fname, ext = op.splitext(op.basename(in_file))
if ext == '.gz':
fname, ext2 = op.splitext(fname)
ext = ext2 + ext
out_file = op.abspath('{}_spmask{}'.format(fname, ext))
out_plot = op.abspath('{}_spmask.pdf'.format(fname))
in_4d_nii = nb.load(in_file)
orientation = nb.aff2axcodes(in_4d_nii.affine)
if in_mask:
mask_data = nb.load(in_mask).get_data()
a = np.where(mask_data != 0)
bbox = np.max(a[0]) - np.min(a[0]), np.max(a[1]) - \
np.min(a[1]), np.max(a[2]) - np.min(a[2])
longest_axis = np.argmax(bbox)
# Input here is a binarized and intersected mask data from previous section
dil_mask = nd.binary_dilation(
mask_data, iterations=int(mask_data.shape[longest_axis] / 9))
rep = list(mask_data.shape)
rep[longest_axis] = -1
new_mask_2d = dil_mask.max(axis=longest_axis).reshape(rep)
rep = [1, 1, 1]
rep[longest_axis] = mask_data.shape[longest_axis]
new_mask_3d = np.logical_not(np.tile(new_mask_2d, rep))
else:
new_mask_3d = np.zeros(in_4d_nii.shape[:3]) == 1
if orientation[0] in ['L', 'R']:
new_mask_3d[0:2, :, :] = True
new_mask_3d[-3:-1, :, :] = True
else:
new_mask_3d[:, 0:2, :] = True
new_mask_3d[:, -3:-1, :] = True
mask_nii = nb.Nifti1Image(new_mask_3d.astype(np.uint8), in_4d_nii.get_affine(),
in_4d_nii.get_header())
mask_nii.to_filename(out_file)
plot_roi(mask_nii, mean_img(in_4d_nii), output_file=out_plot)
return out_file, out_plot
def get_axcodes(self):
"""Get codes for voxel axis derived from affine.
i.e., ('R', 'A', 'S')
"""
if isinstance(self._affine, np.ndarray):
return aff2axcodes(self._affine)
else:
return None
def get_affine_orientation_slice(a):
# get the orientation of the affine, and the slice order
import nibabel as nb
ori=nb.aff2axcodes(a)
if ori[-1] == "I" or ori[-1] == "S":
slc = "AXIAL"
elif ori[-1] == "L" or ori[-1] == "R":
slc="SAGITTAL"
else:
slc="CORONAL"
return ori, slc
def load_nifti(fname, verbose=False):
img = nib.load(fname)
data = img.get_data()
affine = img.get_affine()
if verbose:
print(fname)
print(data.shape)
print(affine)
print(img.get_header().get_zooms()[:3])
print(nib.aff2axcodes(affine))
print
return data, affine
def original_axcodes(self):
"""
axcodes info from the image header
more info: http://nipy.org/nibabel/image_orientation.html
:return: a tuple of axcodes, with each element as axcodes
of an image file
"""
try:
return tuple(nib.aff2axcodes(affine)
for affine in self.original_affine)
except IndexError:
tf.logging.fatal('unknown affine in header %s: %s',
self.file_path, self.original_affine)
raise
def load_nifti(fname, return_img=False, return_voxsize=False,
return_coords=False):
img = nib.load(fname)
data = img.get_data()
vox_size = img.header.get_zooms()[:3]
ret_val = [data, img.affine]
if return_img:
ret_val.append(img)
if return_voxsize:
ret_val.append(vox_size)
if return_coords:
ret_val.append(nib.aff2axcodes(img.affine))
return tuple(ret_val)
def get_header_from_anat(anat_file, hdr={}):
if anat_file is None:
if len(hdr) == 0:
# Defaults
hdr[Header.VOXEL_SIZES] = (0, 0, 0)
hdr[Header.DIMENSIONS] = (1, 1, 1)
return hdr
anat = nib.load(anat_file)
hdr[Header.VOXEL_SIZES] = tuple(anat.get_header().get_zooms())[:3]
hdr[Header.DIMENSIONS] = tuple(anat.get_header().get_data_shape())[:3]
hdr[Header.VOXEL_TO_WORLD] = anat.get_header().get_best_affine()
# We can guess the voxel order from the affine if there is no 0 on the diagonal.
if not np.any(np.diag(hdr[Header.VOXEL_TO_WORLD]) == 0):
hdr[Header.VOXEL_ORDER] = ''.join(nib.aff2axcodes(hdr[Header.VOXEL_TO_WORLD]))
return hdr
affine = img.get_affine()
zooms = img.get_header().get_zooms()[:3]
bvals, bvecs = read_bvals_bvecs(fbvals, fbvecs)
from dipy.core.gradients import gradient_table
gtab = gradient_table(bvals, bvecs, b0_threshold=10)
b0_index = np.where(gtab.b0s_mask == True)[0]
mask = nib.load(fmask).get_data()
print(data.shape)
print(affine)
print(nib.aff2axcodes(affine))
print('>>> Resample data to 1x1x1 mm^3...')
from dipy.align.aniso2iso import resample
data2, affine2 = resample(data, affine,
zooms=zooms,
new_zooms=(1., 1., 1.))
mask2, affine2 = resample(mask, affine,
zooms=zooms,
new_zooms=(1., 1., 1.))
mask2[mask2 > 0] = 1
print(">>> Warp T1 to S0 using ANTS...")
fT1 = join(dname, "T1w_acpc_dc_restore_brain.nii.gz")
fT1_flirt = join(dname, "t1_flirt.nii.gz")
fmat = join(dname, "flirt_affine.mat")
fS0 = join(dname, "dwi_S0_1x1x1.nii.gz")
fFA = join(dname, "dwi_fa_1x1x1.nii.gz")
fT1wS0 = join(dname, "t1_warped_S0.nii.gz")
fdef = join(dname, "MultiVarNew")
img_T1 = nib.load(fT1)
print(img_T1.get_data().shape)
print(img_T1.get_affine())
print(nib.aff2axcodes(img_T1.get_affine()))
del img_T1
flirt_cmd = "flirt -in " + fT1 + " -ref " + fFA + " -out " + fT1_flirt + " -omat " + fmat
print(flirt_cmd)
pipe(flirt_cmd)
br1 = "[" + fS0 + ", " + fT1_flirt + ", 1, 4]"
br2 = "[" + fFA + ", " + fT1_flirt + ", 1.5, 4]"
ants_cmd1 = "ANTS 3 -m CC" + br1 + " -m CC" + br2 + " -o " + fdef + " -i 75x75x10 -r Gauss[3,0] -t SyN[0.25]"
ants_cmd2 = (
"WarpImageMultiTransform 3 "
+ fT1_flirt
+ " "
def main():
np.random.seed(int(time.time()))
parser = buildArgsParser()
args = parser.parse_args()
param = {}
if args.algo not in ["det", "prob"]:
parser.error("--algo has wrong value. See the help (-h).")
if args.basis not in ["mrtrix", "dipy", "fibernav"]:
parser.error("--basis has wrong value. See the help (-h).")
#if np.all([args.nt is None, args.npv is None, args.ns is None]):
# args.npv = 1
if args.theta is not None:
theta = gm.math.radians(args.theta)
elif args.curvature > 0:
theta = get_max_angle_from_curvature(args.curvature, args.step_size)
elif args.algo == 'prob':
theta = gm.math.radians(20)
else:
theta = gm.math.radians(45)
if args.mask_interp == 'nn':
mask_interpolation = 'nearest'
elif args.mask_interp == 'tl':
mask_interpolation = 'trilinear'
else:
parser.error("--mask_interp has wrong value. See the help (-h).")
return
if args.field_interp == 'nn':
field_interpolation = 'nearest'
elif args.field_interp == 'tl':
field_interpolation = 'trilinear'
else:
parser.error("--sh_interp has wrong value. See the help (-h).")
return
param['algo'] = args.algo
param['mask_interp'] = mask_interpolation
param['field_interp'] = field_interpolation
param['theta'] = theta
param['sf_threshold'] = args.sf_threshold
param['sf_threshold_init'] = args.sf_threshold_init
param['step_size'] = args.step_size
param['max_length'] = args.max_length
param['min_length'] = args.min_length
param['is_single_direction'] = False
param['nbr_seeds'] = 0
param['nbr_seeds_voxel'] = 0
param['nbr_streamlines'] = 0
param['max_no_dir'] = int(math.ceil(args.maxL_no_dir / param['step_size']))
param['is_all'] = False
param['isVerbose'] = args.isVerbose
if param['isVerbose']:
logging.basicConfig(level=logging.DEBUG)
if param['isVerbose']:
logging.info('Tractography parameters:\n{0}'.format(param))
if os.path.isfile(args.output_file):
if args.isForce:
logging.info('Overwriting "{0}".'.format(args.output_file))
else:
parser.error(
'"{0}" already exists! Use -f to overwrite it.'
.format(args.output_file))
nib_mask = nib.load(args.mask_file)
mask = BinaryMask(
Dataset(nib_mask, param['mask_interp']))
dataset = Dataset(nib.load(args.sh_file), param['field_interp'])
field = SphericalHarmonicField(
dataset, args.basis, param['sf_threshold'], param['sf_threshold_init'], param['theta'])
if args.algo == 'det':
tracker = deterministicMaximaTracker(field, param['step_size'])
elif args.algo == 'prob':
tracker = probabilisticTracker(field, param['step_size'])
else:
parser.error("--algo has wrong value. See the help (-h).")
return
start = time.time()
# Etienne St-Onge
#load and transfo *** todo test with rotation and scaling
seed_points = np.load(args.seed_points)
seed_dirs = np.load(args.seed_dir)
rotation = nib_mask.get_affine()[:3,:3]
inv_rotation = np.linalg.inv(rotation)
translation = nib_mask.get_affine()[:3,3]
scale = np.array(nib_mask.get_header().get_zooms())
voxel_space = nib.aff2axcodes(nib_mask.get_affine())
print voxel_space
# seed points transfo
# LPS -> voxel_space
if voxel_space[0] != 'L':
print "flip X"
seed_points[:,0] = -seed_points[:,0]
if voxel_space[1] != 'P':
print "flip Y"
seed_points[:,1] = -seed_points[:,1]
if voxel_space[2] != 'S':
print "flip Z"
seed_points[:,2] = -seed_points[:,2]
# other transfo
seed_points = seed_points - translation
seed_points = seed_points.dot(inv_rotation)
seed_points = seed_points * scale
# seed dir transfo
seed_dirs[:,0:2] = -seed_dirs[:,0:2]
seed_dirs = seed_dirs.dot(inv_rotation)
seed_dirs = seed_dirs * scale
if args.inv_seed_dir:
seed_dirs = seed_dirs * -1.0
# Compute tractography
nb_seeds = len(seed_dirs)
if args.test is not None and args.test < nb_seeds:
nb_seeds = args.test
print args.algo," nb seeds: ", nb_seeds
streamlines = []
for i in range(nb_seeds):
s = generate_streamline(tracker, mask, seed_points[i], seed_dirs[i], pft_tracker=None, param=param)
streamlines.append(s)
stdout.write("\r %d%%" % (i*101//nb_seeds))
stdout.flush()
stdout.write("\n done")
stdout.flush()
# transform back
for i in range(len(streamlines)):
streamlines[i] = streamlines[i] / scale
streamlines[i] = streamlines[i].dot(rotation)
streamlines[i] = streamlines[i] + translation
# voxel_space -> LPS
if voxel_space[0] != 'L':
streamlines[i][:,0] = -streamlines[i][:,0]
if voxel_space[1] != 'P':
streamlines[i][:,1] = -streamlines[i][:,1]
if voxel_space[2] != 'S':
streamlines[i][:,2] = -streamlines[i][:,2]
lines_polydata = lines_to_vtk_polydata(streamlines, None, np.float32)
save_polydata(lines_polydata, args.output_file , True)
lengths = [len(s) for s in streamlines]
if nb_seeds > 0:
ave_length = (sum(lengths) / nb_seeds) * param['step_size']
else:
ave_length = 0
str_ave_length = "%.2f" % ave_length
str_time = "%.2f" % (time.time() - start)
print(str(nb_seeds) + " streamlines, with an average length of " +
str_ave_length + " mm, done in " + str_time + " seconds.")
def get_nifti_voxel_space(nii):
return nib.aff2axcodes(nii.get_affine())
def main():
parser = buildArgsParser()
args = parser.parse_args()
param = {}
if args.pft_theta is None and args.pft_curvature is None:
args.pft_theta = 20
if not np.any([args.nt, args.npv, args.ns]):
args.npv = 1
if args.theta is not None:
theta = gm.math.radians(args.theta)
elif args.curvature > 0:
theta = get_max_angle_from_curvature(args.curvature, args.step_size)
elif args.algo == 'prob':
theta = gm.math.radians(20)
else:
theta = gm.math.radians(45)
if args.pft_curvature is not None:
pft_theta = get_max_angle_from_curvature(args.pft_curvature, args.step_size)
else:
pft_theta = gm.math.radians(args.pft_theta)
if args.mask_interp == 'nn':
mask_interpolation = 'nearest'
elif args.mask_interp == 'tl':
mask_interpolation = 'trilinear'
else:
parser.error("--mask_interp has wrong value. See the help (-h).")
return
if args.field_interp == 'nn':
field_interpolation = 'nearest'
elif args.field_interp == 'tl':
field_interpolation = 'trilinear'
else:
parser.error("--sh_interp has wrong value. See the help (-h).")
return
param['random'] = args.random
param['skip'] = args.skip
param['algo'] = args.algo
param['mask_interp'] = mask_interpolation
param['field_interp'] = field_interpolation
param['theta'] = theta
param['sf_threshold'] = args.sf_threshold
param['pft_sf_threshold'] = args.pft_sf_threshold if args.pft_sf_threshold is not None else args.sf_threshold
param['sf_threshold_init'] = args.sf_threshold_init
param['step_size'] = args.step_size
param['max_length'] = args.max_length
param['min_length'] = args.min_length
param['is_single_direction'] = args.is_single_direction
param['nbr_seeds'] = args.nt if args.nt is not None else 0
param['nbr_seeds_voxel'] = args.npv if args.npv is not None else 0
param['nbr_streamlines'] = args.ns if args.ns is not None else 0
param['max_no_dir'] = int(math.ceil(args.maxL_no_dir / param['step_size']))
param['is_all'] = args.is_all
param['is_act'] = args.is_act
param['theta_pft'] = pft_theta
if args.not_is_pft:
param['nbr_particles'] = 0
param['back_tracking'] = 0
param['front_tracking'] = 0
else:
param['nbr_particles'] = args.nbr_particles
param['back_tracking'] = int(
math.ceil(args.back_tracking / args.step_size))
param['front_tracking'] = int(
math.ceil(args.front_tracking / args.step_size))
param['nbr_iter'] = param['back_tracking'] + param['front_tracking']
param['mmap_mode'] = None if args.isLoadData else 'r'
if args.isVerbose:
logging.basicConfig(level=logging.DEBUG)
logging.debug('Tractography parameters:\n{0}'.format(param))
if os.path.isfile(args.output_file):
if args.isForce:
logging.info('Overwriting "{0}".'.format(args.output_file))
else:
parser.error(
'"{0}" already exists! Use -f to overwrite it.'
.format(args.output_file))
include_dataset = Dataset(
nib.load(args.map_include_file), param['mask_interp'])
exclude_dataset = Dataset(
nib.load(args.map_exclude_file), param['mask_interp'])
if param['is_act']:
mask = ACT(include_dataset, exclude_dataset,
param['step_size'] / include_dataset.size[0])
else:
mask = CMC(include_dataset, exclude_dataset,
param['step_size'] / include_dataset.size[0])
dataset = Dataset(nib.load(args.sh_file), param['field_interp'])
field = SphericalHarmonicField(
dataset, args.basis, param['sf_threshold'],
param['sf_threshold_init'], param['theta'])
if args.algo == 'det':
tracker = deterministicMaximaTracker(field, param['step_size'])
elif args.algo == 'prob':
tracker = probabilisticTracker(field, param['step_size'])
else:
parser.error("--algo has wrong value. See the help (-h).")
return
pft_field = SphericalHarmonicField(
dataset, args.basis, param['pft_sf_threshold'],
param['sf_threshold_init'], param['theta_pft'])
pft_tracker = probabilisticTracker(pft_field, param['step_size'])
# ADD Seed input
# modify ESO
nib_mask = nib.load(args.map_include_file)
seed_points = np.load(args.seed_points)
seed_dirs = np.load(args.seed_dir)
rotation = nib_mask.get_affine()[:3,:3]
inv_rotation = np.linalg.inv(rotation)
translation = nib_mask.get_affine()[:3,3]
scale = np.array(nib_mask.get_header().get_zooms())
voxel_space = nib.aff2axcodes(nib_mask.get_affine())
print voxel_space
# seed points transfo
# LPS -> voxel_space
print scale
if voxel_space[0] != 'L':
print "flip X"
seed_points[:,0] = -seed_points[:,0]
if voxel_space[1] != 'P':
print "flip Y"
seed_points[:,1] = -seed_points[:,1]
if voxel_space[2] != 'S':
print "flip Z"
seed_points[:,2] = -seed_points[:,2]
# other transfo
seed_points = seed_points - translation
seed_points = seed_points.dot(inv_rotation)
seed_points = seed_points * scale
# seed dir transfo
seed_dirs[:,0:2] = -seed_dirs[:,0:2]
seed_dirs = seed_dirs.dot(inv_rotation)
seed_dirs = seed_dirs * scale
if args.inv_seed_dir:
seed_dirs = seed_dirs * -1.0
# Compute tractography
nb_seeds = len(seed_dirs)
if args.test is not None and args.test < nb_seeds:
nb_seeds = args.test
# end modify ESO
# tracker to modify
# modify ESO
start = time.time()
streamlines = []
for i in range(nb_seeds):
s = generate_streamline(tracker, mask, seed_points[i], seed_dirs[i], pft_tracker=pft_tracker, param=param)
streamlines.append(s)
stdout.write("\r %d%%" % (i*101//nb_seeds))
stdout.flush()
stdout.write("\n done")
stdout.flush()
stop = time.time()
# end modify ESO
# ADD save fiber output
# modify ESO
for i in range(len(streamlines)):
streamlines[i] = streamlines[i] / scale
streamlines[i] = streamlines[i].dot(rotation)
streamlines[i] = streamlines[i] + translation
# voxel_space -> LPS
if voxel_space[0] != 'L':
streamlines[i][:,0] = -streamlines[i][:,0]
if voxel_space[1] != 'P':
streamlines[i][:,1] = -streamlines[i][:,1]
if voxel_space[2] != 'S':
streamlines[i][:,2] = -streamlines[i][:,2]
lines_polydata = lines_to_vtk_polydata(streamlines, None, np.float32)
save_polydata(lines_polydata, args.output_file , True)
# end modify ESO
lengths = [len(s) for s in streamlines]
if nb_seeds > 0:
ave_length = (sum(lengths) / nb_seeds) * param['step_size']
else:
ave_length = 0
str_ave_length = "%.2f" % ave_length
str_time = "%.2f" % (stop - start)
print(str(nb_seeds) + " streamlines, with an average length of " +
str_ave_length + " mm, done in " + str_time + " seconds.")
# load volume
mask_file = args.volume
volume_nib = nib.load(mask_file)
# load tracto
init_streamlines_list = []
for filename in args.fibers:
init_streamlines_list.append(get_streamlines(load_streamlines_poyldata(filename)))
print filename, len(init_streamlines_list[-1])
# Transform tracto to Voxel space
rotation = volume_nib.get_affine()[:3,:3]
inv_rotation = np.linalg.inv(rotation)
translation = volume_nib.get_affine()[:3,3]
scale = np.array(volume_nib.get_header().get_zooms())
voxel_space = nib.aff2axcodes(volume_nib.get_affine())
print voxel_space
# seed points transfo
# LPS -> voxel_space
vertices_list = []
dirs_list = []
for streamlines in init_streamlines_list:
for streamline in streamlines:
if voxel_space[0] != 'L':
#print "flip X"
streamline[:,0] = -streamline[:,0]
if voxel_space[1] != 'P':
#print "flip Y"
streamline[:,1] = -streamline[:,1]
if voxel_space[2] != 'S':
parser.add_argument('-tx', type=float, default=None, help='x translation')
parser.add_argument('-ty', type=float, default=None, help='y translation')
parser.add_argument('-tz', type=float, default=None, help='z translation')
args = parser.parse_args()
# get transform
nib_mask = nib.load(args.mask)
lines = get_streamlines(load_streamlines_poyldata(args.tract))
rotation = nib_mask.get_affine()[:3,:3]
inv_rotation = np.linalg.inv(rotation)
translation = nib_mask.get_affine()[:3,3]
scale = np.array(nib_mask.get_header().get_zooms())
voxel_space = nib.aff2axcodes(nib_mask.get_affine())
shape = nib_mask.get_data().shape
print shape
# transform
if not args.no_transfo:
if args.lps_ras:
print "LPS -> RAS"
print "Not implemented"
raise NotImplementedError()
else:
print "LPS ->", voxel_space, " mm"
for i in range(len(lines)):
if voxel_space[0] != 'L':
lines[i][:,0] = -lines[i][:,0]